1. Field of the Invention
The present invention generally relates to systems and methods for characterizing a polishing process. Certain embodiments relate to systems and methods for evaluating optical and/or eddy current data obtained during polishing of a specimen to determine a characteristic of the polishing process.
2. Description of the Related Art
The following descriptions and examples are not admitted to be prior art by virtue of their inclusion within this section.
Fabricating semiconductor devices such as logic and memory devices may typically include processing a specimen such as a semiconductor wafer using a number of semiconductor fabrication processes to form various features and multiple levels of the semiconductor devices. For example, insulating (or dielectric) materials may be formed on multiple levels of a substrate using deposition processes such as chemical vapor deposition (“CVD”), physical vapor deposition (“PVD”), and atomic layer deposition (“ALD”). Such insulating materials may electrically isolate conductive structures of a semiconductor device formed on the substrate. For example, the insulating materials may be used to form an interlevel dielectric or shallow trench isolation regions. Conductive materials may also be formed on a substrate using the deposition processes described above. In addition, conductive materials may also be formed on a substrate using a plating process. Chemical-mechanical polishing (“CMP”) may typically be used in the semiconductor industry to reduce elevational disparities or to planarize layers of such materials on a specimen. Additional examples of semiconductor fabrication processes may include, but are not limited to lithography, etch, ion implantation, and cleaning. Multiple semiconductor devices may be fabricated in an arrangement on a semiconductor wafer and then separated into individual semiconductor devices.
Characterizing, monitoring, and/or controlling such semiconductor fabrication processes is an important aspect of semiconductor device manufacturing. A number of techniques are presently available for such characterizing, monitoring, and/or controlling. For example, one presently available method to control a CMP process for shallow trench isolation is a polishing-time based method, which uses a fixed polishing time determined from polishing results of test, or monitor, wafers. In situ end point detection methods based on motor current and carrier vibration techniques are also currently available. In addition, post-CMP in-tool film thickness measurements are currently used.
There are, however, several disadvantages to such currently available methods for characterizing, monitoring, and/or controlling a CMP process. For example, in a CMP process, many variable parameters such as pad condition, slurry chemistry, incoming wafer film thickness, and circuit pattern density may affect the required polishing time. The polishing-time based method may not effectively handle these changes in the polishing conditions, and thus often produces over-polished or under-polished results. In addition, measuring monitor wafers reduces production throughput and thus overall equipment efficiency. Motor current and carrier vibration endpoint detection methods may not provide planarization information in different wafers areas and may not be effective for a shallow trench isolation (STI) process.
Currently available methods for characterizing, monitoring, and/or controlling a CMP process may also include ex situ and in situ endpoint detection methods. Ex situ methods include analyzing the wafer surface after a polishing process has finished. For example, such analyzing may include removing the wafer from the polishing chamber and loading the wafer in a metrology system. In situ methods include indirect methods such as slurry byproduct monitoring and methods described above such as motor current monitoring and carrier head vibration monitoring. One currently available in situ direct method uses an eddy current-based proximity sensor. The eddy current sensor provides a relative indication of thick metal films such as copper by sensing only the in-phase component of the induced eddy current.
There are also, however, several disadvantages to currently available ex situ methods for characterizing, monitoring, and/or controlling a CMP process. For example, CMP tool throughput may be reduced due to ex situ endpointing systems because the wafer must be removed from the process tool, analyzed, and marginalities of its polishing must be resolved before the next wafer can be polished. Ex situ methods are also more problematic due to the difficulty of resuming CMP processing of a wafer that is under-polished. Furthermore, ex situ methods are even more problematic because over-polishing of wafers cannot be actively prevented, only reported after the fact. Therefore, ex situ process control methods may suffer from a high scrapped wafer rate.
In addition, there are several disadvantages to currently available in situ methods for characterizing, monitoring, and/or controlling a CMP process. For example, in situ, indirect methods provide no local information on films. Therefore, local information often has to be determined by ex situ spot checking of the wafers. Moreover, indirect monitoring makes process tuning more difficult. In addition, indirect methods are feasible only with certain polishing pads, slurries, speeds, and pressure settings. Therefore, these constraints limit the options for CMP processes. Sometimes such constraints may translate into diminished throughput and polish quality. Currently available in situ direct methods that use eddy current-based sensors but report only a relative thickness value are known in the art, but a relative process variable is difficult to incorporate into a recipe for transport between process tools. Moreover, these devices do not compensate for temperature changes that may affect the sensor output.
Currently available methods for whole-wafer measurements of thickness, typically, do not provide spatial resolution. For example, some currently available methods use a fixed sensor such as a sensor mounted on a shaft of a table supporting the wafer. Therefore, such sensors can only measure one location of the wafer (i.e., the center spot). Such methods may provide relatively poor performance because the entire wafer does not polish at the same rate as the observed spot.
In rotary platen/rotary carrier machines, sensors may be fixed off-center under the platen to sweep the wafer as the table rotates. Depending upon the ratio of the rotational speeds of the platen and the carrier, the sensor path over the wafer may be different with each sweep. Such methods process the measurements within annular zones on the wafer. Therefore, although such methods correlate the measurements to a radial location with respect to the wafer center, the measurements are not correlated to an angular location. As such, these techniques provide no method by which to associate a specific spatial location on the wafer with a specific measurement. For example, data processing on a control computer may indicate that a certain zone was polished too long. This means that CMP defects such as dishing and erosion are likely to be present in this annular zone. The data processing, however, does not determine where this region lies, except that it is a given distance from the wafer center. Therefore, annular-zone based measurements provide limited spatial resolution based on the sensor's distance from the wafer center. Examples of such methods are illustrated in U.S. Pat. No. 5,893,796 to Birang et al., U.S. Pat. No. 5,964,643 to Birang et al., U.S. Pat. No. 6,045,439 to Birang et al., U.S. Pat. No. 6,159,073 to Wiswesser et al., and U.S. Pat. No. 6,280,290 to Birang et al., which are incorporated by reference as if fully set forth herein.
In some CMP system configurations, such information may be passed to another control computer which continues the wafer planarization on another platen with different process parameters. However, the annular zone-based information may not be useful since the angular orientation of the wafer is lost in the transfer to the platen used in the second step. A program of the second control computer may regenerate a full wafer map of surface film features on the wafer, but in the time required to regenerate the map, the wafer may be damaged by over-polishing while these complicated algorithms execute.